{
  "id": 687312,
  "title": "Why Is My Unselected Notebook Scoring Higher Than Top 1 on Private?",
  "url": "/competitions/stanford-rna-3d-folding-2/discussion/687312",
  "author_name": "Van-Phuc Huynh",
  "post_date": "2026-04-02T20:05:58.265000",
  "votes": 0,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I’m confused — my unselected notebook shows a higher private score than the leaderboard top. Can someone explain what’s happening with the 0.53189 private score?</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2106526%2F1039f46dc9387719a2325811a1606fe4%2FCapture.PNG?generation=1775160169852090&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": 3434334,
      "postDate": "2026-04-02T20:28:19.503Z",
      "content": "<p>It is definitely confusing, and we'll look at ways to make things more clear. But let me explain how things work.</p>\n<ol>\n<li>During the competition, the private test set had a subset of targets; it wasn't the actual final dataset.</li>\n<li>At the end of the competition, the final rerun dataset was uploaded</li>\n<li>Selected notebooks were rerun against the rerun Test dataset</li>\n<li>Variability in the Public score is due to the stochastic nature of most algorithms</li>\n<li>Variability in the Private score is due to a rescore on the full set of observations</li>\n</ol>",
      "rawMarkdown": "It is definitely confusing, and we'll look at ways to make things more clear. But let me explain how things work.\n1. During the competition, the private test set had a subset of targets; it wasn't the actual final dataset.\n2. At the end of the competition, the final rerun dataset was uploaded\n3. Selected notebooks were rerun against the rerun Test dataset\n4. Variability in the Public score is due to the stochastic nature of most algorithms\n5. Variability in the Private score is due to a rescore on the full set of observations",
      "replies": [
        {
          "id": 3434345,
          "postDate": "2026-04-02T20:46:32.243Z",
          "content": "<p>Thanks for the clarification. It makes sense that stochasticity and the rerun on the full dataset can lead to noticeable differences in scores.</p>",
          "rawMarkdown": "Thanks for the clarification. It makes sense that stochasticity and the rerun on the full dataset can lead to noticeable differences in scores."
        }
      ]
    },
    {
      "id": 3434319,
      "postDate": "2026-04-02T20:05:58.267Z",
      "content": "<p>I’m confused — my unselected notebook shows a higher private score than the leaderboard top. Can someone explain what’s happening with the 0.53189 private score?</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2106526%2F1039f46dc9387719a2325811a1606fe4%2FCapture.PNG?generation=1775160169852090&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "I’m confused — my unselected notebook shows a higher private score than the leaderboard top. Can someone explain what’s happening with the 0.53189 private score?\n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2106526%2F1039f46dc9387719a2325811a1606fe4%2FCapture.PNG?generation=1775160169852090&alt=media)"
    }
  ],
  "comments": [
    {
      "id": 3434334,
      "author_name": "inversion",
      "author_url": "",
      "post_date": "2026-04-02T20:28:19.503000",
      "content": "<p>It is definitely confusing, and we'll look at ways to make things more clear. But let me explain how things work.</p>\n<ol>\n<li>During the competition, the private test set had a subset of targets; it wasn't the actual final dataset.</li>\n<li>At the end of the competition, the final rerun dataset was uploaded</li>\n<li>Selected notebooks were rerun against the rerun Test dataset</li>\n<li>Variability in the Public score is due to the stochastic nature of most algorithms</li>\n<li>Variability in the Private score is due to a rescore on the full set of observations</li>\n</ol>",
      "votes": 0,
      "replies": [
        {
          "id": 3434345,
          "author_name": "Van-Phuc Huynh",
          "author_url": "",
          "post_date": "2026-04-02T20:46:32.243000",
          "content": "<p>Thanks for the clarification. It makes sense that stochasticity and the rerun on the full dataset can lead to noticeable differences in scores.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3434334": "It is definitely confusing, and we'll look at ways to make things more clear. But let me explain how things work.\n1. During the competition, the private test set had a subset of targets; it wasn't the actual final dataset.\n2. At the end of the competition, the final rerun dataset was uploaded\n3. Selected notebooks were rerun against the rerun Test dataset\n4. Variability in the Public score is due to the stochastic nature of most algorithms\n5. Variability in the Private score is due to a rescore on the full set of observations",
    "3434319": "I’m confused — my unselected notebook shows a higher private score than the leaderboard top. Can someone explain what’s happening with the 0.53189 private score?\n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2106526%2F1039f46dc9387719a2325811a1606fe4%2FCapture.PNG?generation=1775160169852090&alt=media)"
  }
}